fMRIPrep visual reports

Author

Anastasios Dadiotis

Error reports

  • First check for the error reports at the very end of the html file

Summary

Overall summary of the pipeline: information about anatomical and functional images

This should much the options we defined on the script that we used to run fmriprep

Anatomical Section

Brain mask and brain tissue segmentation of the T1w

  • A good overview of the surfaces as well as the brain masks created by Freesurfer
    • White matter boundary is outlined in blue
    • Estimated brain mask is outlined in red
    • Grey matter boundary is outlined in magenda

Example:

Inaccurate brain mask

  • The brain mask should closely follow the contour of the brain.
  • An inaccurate brain mask presents “bumps” surrounding high-intensity areas of signal outside of the cortex (e.g., a mask inclusding patches of he skull) and/or holes surrounding signal drop-out regions.

Error in brain tissue segmentation of T1w images

  • To confirm the good quality of the segmentation verify that:
    • The magenda contour outline the ventricles
    • The blue contour followed the boundary between the white matter and the grey matter
  • Exclusion criteria: inclusion of tissue other than the tissue of interest in the contour delineations

Example:

Spatial normalization of the anatomical T1w reference

  • A GIF in which if you hover your mouse over the image, the image will flicker between participant space and to the chosen template (e.g. here MNI152NLin6Asym space).
  • This can show you how well the native space was registered to the standard space overall
  • Failure in normalization to MNI space
    • Normalization must me perfect
  • To verify successful normalization assess the correct alignment of the following structures (in order of importance):
  1. Ventricles
  2. Subcortical regions
  3. Corpus Callosum
  4. Cerebellum
  5. Cortical Grey Matter (GM)
  • Misalignment of 1,2 or 3: immediate exclusion
  • For 5 a bit more loose because volumetric (image) registration may not resolve substantial inter-individual differences
  • Any extreme stretching or distortion of the T1w image also indicates a failed normalization

Tip: Put your mouse at the edge of the participant brain, should be perfect when flickers1 between the 2 spaces (This is not the case for the MRIQC output that has only the 1st iteration while here the algorithm has multiple iteration and trying to optimize the procedure).

Tip: Make sure to check the alignment not only in the outlines of the brain (see above) but also in the internal structures such as the ventricles.

Tip: (From discussion with Gaëlle) Do not pay so much emphasis in image x=0 as it it difficult and errors must be expected

Example:

Surface reconstruction

  • White matter in blue (should follow the boundary between the grey matter and the white matter)
  • Pial surface in red (should outline the outer boundary of the grey matter and exclude the cerebellum)
  • These are overlaid on the participant’s T1w image
  • Usually you do not exclude data here except the reconstructed surfaces are extremely inaccurate but you should already have seen that in MRIQC

From here they suggest also:

  • That you check the same checks for the mask and segmentation

  • And they provide this image as a good example:

Functional Section

From here: Look at the information text provided to check:

  • Is there any mismatching parameters(TR, phase encoding direction, sequence details, slice timing, susceptibility correction,registration)?

  • Is there any non-steady state volumes? Make sure you note them for later on adding to your GLM analysis ( columns indicate non-steady state volumes with a single 1 value and 0 elsewhere).

Alignment between the anatomical reference of the fieldmap and the target EPI

  • After discussion with Gaëlle patterns like the one’s highlighted below are normal

Susceptibility distortion correction

  • A GIF that shows before and after distortion correction
  • Note that the correction is affected by the direction AP vs PA
  • Any observation of susceptibility distortion artifacts leads to the exclusion of the scan
  • The after image should also appear less distorted and shaped more like a normal brain, this may be subtle when the distortion correction is working, but can really, really stand out when it fails.

Example:

Brain mask and (temporal/anatomical) CompCor ROIs

  • Data driven approach to add nuisance regressors
  • If you want to use this approach check also the correlation heatmap
  • Make sure the brain mask shown by the red contour is outside the brain mask in the functional images
  • Make sure the magenda lines are well inside the white matter/csf
  • In general, areas outlined by the blue lines should be areas with high CSF or blood flow, such as
    • between the hemispheres
    • in ventricles
    • between the cortex and cerebellum.
  • These are the most variable voxels that will be used later on for functional component correction

Examples:

  • This one is worth checking, as the red line cuts of some of the spinal cord at the bottom. Ultimately, this is usable:

  • The distortion in the cerebellum is worth noting, but this is usable as well:

  • The red line includes some dura at x=-2. Since it’s only on the midline, it’s still acceptable:

  • Severe dropout like this isn’t usable:

  • Bad

Alignment of functional and anatomical MRI data (coregistration)

  • Co - registration problem

    • Check the alignment of image intensity edges and the anatomical landmarks (e.g. the ventricles and the corpus callosum) between the BOLD and the T1w images

Time-series plots and carpet plot

  • Contains time series plots for

    • GS: Global Signal

    • GSCSF: Global Signal of the Cerebral Spinal Fluid

    • GSWM: Global Signal of the White Matter

  • and 2 measures of motion

    • DVARS
    • FD: Framewise displacement
  • Changes in motion tend to be correlated with GS and it is up to us to decide if we include any of these vars in the model. In general DVARS and FD are good ways to account for signal caused by motion artifacts

  • Look for any big spikes in any of the line plots.

  • For carpet plot look for any columns that all seem to have a jump in values, this will look like vertical bands or lines down the plot covers across the entire column for that time point.

Correlations among nuisance regressors

  • High correlations can maybe explained by a motion that caused a signal change in one tissue type that affects the other.

  • Ideally all these components should be capturing the different elements, noise in your data. However if you see that every single components correlate with the global signal in a high level then,it means something might be wrong and needs further investigation with the data.

Smoothing

Important Note: Smoothing is not in the pipeline of fmriprep, therefore it is a step that has to be executed after running the fmriprep pipeline.

  • Smoothing has been ommited by design. The fmriprep does not make assumption on how you are going to analyze your data

    • e.g. Some MVPA studies do not do any smoothing at all before the first level analysis
  • The amount of smoothing is a matter of judgement

    • Experiments that are focused on larger cortical areas, probably want to use larger smoothing kernels

Tip: Since we have multiple tasks with different goals, smoothing should be executed with this in mind and probably individually for each task/goal. When performing that use a prefix (e.g. smo0.8 or sth along those lines) in the name of the data so it is quite clear and easy to find what was the smoothing!

ToDO